Fifty-seventh annual meeting of the American association of physicists in medicine
SU-C-303-07: Influence of Image Registration Algorithms and Noise Levels On the Accuracy of Fractional Regional Ventilation
The derivation of fractional regional lung-ventilation (FRV) involves registering images corresponding to extreme phases of the respiratory cycle. The purpose of this work was to investigate the influence of registration algorithms and image noise on the accuracy of the mouse lung tidal volume (TV) by employing a density-based mass-conservation technique in the calculation of fractional regional lung ventilation (FRV).
Three mice mechanically ventilated using a fixed volume (0.4 ml) ventilator. CT images were acquired at full-inhale and end-exhale breath hold. Deformable image registration between inhale and exhale images was performed using three image registration algorithms: B-splines, Demons-Diffusion and Demons-Fluid model. The impact of noise was studied by adding Gaussian noise resulting in images with relative signal-to-noise ratio (SNR) of 1 (zero noise), 0.65 (relative to zero noise), 0.39 and 0.26. A density-based technique for evaluation of FRV from CT images was used to calculate tidal volume. The global TV calculated from the FRV maps was benchmarked to the known ventilated mouse lung volume of 0.4 ml.
The TV obtained using the B-splines, Demons-Diffusion and Demons-Fluid registration algorithms with zero noise (relative SNR=1) were 0.28±0.03 ml, 0.22±0.04 ml and 0.41±0.01 ml, respectively. We observed a decline in the calculated tidal volume as SNR decreased. The greatest decline (27%) was observed with the Demons-Fluid algorithm. However, the absolute value of the calculated TV with a relative SNR=0.26 was 0.30±0.05 ml, still the most accurate TV of the three registration algorithms investigated. The B-splines and Demons-Diffusion algorithm were less sensitive to noise and resulted in TV declines of 18% with a relative SNR=0.26.
The Demons-Fluid image registration algorithm consistently provided the most accurate calculated absolute TV. Although it was the most sensitive to image noise, the calculated absolute TV was more accurate than the B-splines and Demons Diffusion algorithms.